Update README.md
Browse files
README.md
CHANGED
|
@@ -25,6 +25,8 @@ context 32k, cache 8: 70.3GiB (fits in 3x 3090)
|
|
| 25 |
context 32k, cache 16: 78.7GiB (fits in A100 80GB)
|
| 26 |
|
| 27 |
# Super epic scientific test results
|
| 28 |
-
The 2.65bpw version suffered greatly, it's not completely broken, but it's not good either.
|
| 29 |
-
The 4.35bpw version is worse than normal 4k goliath but better than goliath with rope scale applied for 8k+ context.
|
| 30 |
-
The version using the PIPPA dataset produces worse results than the one using the default dataset on any context length.
|
|
|
|
|
|
|
|
|
| 25 |
context 32k, cache 16: 78.7GiB (fits in A100 80GB)
|
| 26 |
|
| 27 |
# Super epic scientific test results
|
| 28 |
+
- The 2.65bpw version suffered greatly, it's not completely broken, but it's not good either.
|
| 29 |
+
- The 4.35bpw version is worse than normal 4k goliath but better than goliath with rope scale applied for 8k+ context.
|
| 30 |
+
- The version using the PIPPA dataset produces worse results than the one using the default dataset on any context length.
|
| 31 |
+
|
| 32 |
+
My current strategy is to use the original goliath until its context is full and then switch over to this one.
|